- Leadership Journey
- Posts
- Lisa Su’s $200 B Chip Comeback
Lisa Su’s $200 B Chip Comeback
How a quiet engineer pulled AMD back from the brink and rewired the race for high performance computing
On 31 July 2025 Advanced Micro Devices (AMD) reported record quarterly revenue of $9.4 billion, up 21 percent year‑on‑year, driven by blockbuster demand for its MI300 AI accelerators. Moments later CEO Lisa Su told analysts, “We bet the company on high‑performance leadership and we’re still just getting started.” A decade ago AMD hovered near penny‑stock territory; today its market cap flirts with $200 billion, proof that long‑game engineering can outpace deeper pockets.
Origin Moment: Soldering Irons & Supercomputers
Lisa Su was born in Tainan, Taiwan, in 1969 and moved to New York at age three. Weekends were spent dismantling radios with her father, an electrical engineer, and at 12 she bought her first soldering iron. By 16 she was interning at Analog Devices. Three MIT degrees later including a PhD on silicon‑on‑insulator transistors she joined IBM’s semiconductor R&D labs and published dozens of papers on chip materials.
The career pivot came in 2012 when cash‑strapped AMD recruited her as SVP of global business units. The company was bleeding money after a series of Bulldozer CPU misfires. Su accepted on one condition: full access to the roadmap so she could “fix the architecture, not the marketing.”
First Turning Point: The Zen Gamble (2014‑2017)
Appointed CEO in October 2014, Su cancelled low‑margin product lines and funneled scarce R&D dollars into a new CPU core called Zen. Engineers had 36 months and a shoestring budget to beat Intel’s performance‑per‑watt. Su ran bi‑weekly “War Room” reviews, limiting slide decks to three data‑dense pages.
March 2017: Ryzen desktops launched, matching Intel’s flagship chips at lower prices. AMD’s CPU market share jumped from 8 percent to 17 percent within a year.
Why it mattered: Singular focus on an architectural leap rather than incremental tweaks reset competitive dynamics and restored engineering pride.
Cultural Reset: High Accountability, Zero Politics
Su’s mantra, “Best idea wins,” reshaped AMD’s culture. Org charts were flattened, opaque program codes replaced by plain‑language milestones, and weekly cross‑team code reviews became standard. Bonus pools now hinge on silicon yield and performance metrics, not head‑count size.
Employees cite the “Say Red, Show Data” ritual: every meeting starts by stating the riskiest assumption (red flag) followed by supporting data. The practice surfaces issues early and discourages sandbagging. Attrition among senior architects fell below 5 percent by 2022, an industry low.
Second Turning Point: Betting on AI Accelerators (2020‑2025)
NVIDIA’s dominance in AI could have trapped AMD in perpetual second place. Instead, Su green‑lit the MI200 series GPUs tuned for massive matrix math despite pandemic supply shocks. She secured 6 nm capacity at TSMC three years ahead and lured Meta and Microsoft into co‑design agreements.
Q2 2025: The MI300X shipped to hyperscalers, delivering 1.5× better FLOPS‑per‑watt than NVIDIA’s H100 in independent tests. AI/data‑center revenue now outstrips gaming, and AMD’s share of AI‑training hardware has climbed to 22 percent.
Key insight: Controlled parity is not enough; selective over‑performance creates perception flywheels that re‑price the whole portfolio.
Mindset & Habits: Four Practices You Can Steal
Habit | What Su Does | Why It Works |
Lab Walk Tuesdays | Spends three hours each week on the fab floor and simulation pods. | Keeps leadership grounded in physics, not PowerPoint. |
Three‑Slide Rule | Decision briefs limited to three data‑packed slides. | Forces clarity and speeds iteration. |
Risk‑First Stand‑ups | Teams open with worst‑case failure, then mitigation plan. | Surfaces hidden blockers early. |
Mentor‑Shadow Swaps | Senior architects pair with junior recruits on six‑week rotations. | Transfers tacit knowledge and boosts retention. |
Sunday Patent Reads | Reviews one new semiconductor patent each weekend. | Sparks cross‑domain ideas for future roadmaps. |
Lessons for Readers
1. Bet the Company, Not the Quarter
Zen consumed 80 percent of R&D when cash was tight; bold allocation resets trajectories faster than incrementalism. Massive swings in focus also broadcast, inside and out, that incremental fixes won’t solve structural problems.
2. Architect Culture Like Chips
Rituals such as “Say Red, Show Data” hard‑wire transparency and speed culture should be engineered, not inherited. Iterating on behaviors the way engineers iterate on silicon creates repeatable pathways for innovation.
3. Chase Performance‑per‑Watt
A single north‑star metric aligns hardware, firmware, and marketing, shrinking debate cycles. Concentrating on one quantifiable target galvanizes teams and streamlines trade‑off decisions.
4. Secure Capacity Early
Pre‑ordering TSMC capacity during global shortages locked in advantage; foresight beats spot‑market panic. Securing long‑lead resources de‑risks execution and can force rivals into costly last‑minute bids.
5. Teach & Retain
Mentor‑shadow rotations prevent brain drain and multiply expertise critical when talent wars spike. Knowledge‑sharing structures also forge loyalty, turning veterans into force‑multipliers for new talent.
Weekly Challenge
Identify the riskiest assumption in your current project. Open your next team meeting by stating it plainly and asking for data that confirms or challenges it. Send us a 100‑word recap, and we’ll feature the boldest “Say Red, Show Data” story next Thursday.